Trust
Verify us in about a minute.
Nothing on this page asks for belief. Four moves, each one checkable from where you sit: run the benchmarks live, diff the deterministic receipts, read the measurement doctrine, and ask the engine what it serves.
Run both public benchmarks, right here.
The runs below execute against the same serving pack that answers every client match. No key, nothing special-cased for the test. The wrong-entity rate prints first because a confident wrong answer is the one that costs you.
Diff the receipt.
Every run returns a sha256 over the set version, the reference pack, and every row disposition. Identical inputs produce the identical receipt on any deploy, so run it twice and compare, or compare against ours. Two parties can agree on a number without trusting each other.
Read how the numbers are made.
The methodology page explains the two benchmarks, the per-tier calibration program with its labeled sample sizes, and the leak-free discipline every eval clears before its number counts for anything.
Ask the engine directly.
The capabilities endpoint answers anonymously with what the engine serves right now. Agents read it before every session, and you can read it in a terminal.
curl -s https://turntodata.ai/api/capabilitiesRun both public benchmarks yourself, live against the serving pack. No key needed. A representative CRM file shows field performance, and an adversarial set shows the honest floor. Results are cached per reference pack, so every run returns the identical receipts.
Both sets run in your browser against the serving pack. The receipts and per-row dispositions come back with the numbers.
The standing posture
What holds on every page.
If something looks wrong
Found a number that does not hold up?
Tell us. A benchmark that only ships when we win is marketing, and we run the other kind. Security reports go to the security.txt contact, everything else to the address on the about page.